By AIM · AIM Media House
AI’s growth is colliding with electricity limits. Data center operators are finding that powering ever-larger “cluster farms” in single locations is becoming unfeasible.
In a move that may help alleviate this barrier, Cisco just introduced the Silicon One P200 chip and 8223 router, which are designed to enable AI workloads to be distributed across multiple data centers.
Cisco says the 8223, powered by P200, delivers 51.2 Tbps throughput while cutting power consumption by about 65%compared to its previous generation systems. The system supports 64 ports of 800G coherent optics and is capable of connecting data centers separated by distances up to 1,000 km.
Cisco says the 8223, powered by P200, delivers 51.2 Tbps throughput while cutting power consumption by about 65%compared to its previous generation systems. The system supports 64 ports of 800G coherent optics and is capable of connecting data centers separated by distances up to 1,000 km.
https://twitter.com/Grit_Capital/status/1975934881411535016 Martin Lund , Executive Vice President of Cisco’s Common Hardware Group, said , “AI compute is outgrowing the capacity of even the largest data center, driving the need for reliable, secure connection of data centers hundreds of miles apart.” Lund added that AI firms are placing data centers “wherever you can get power.” Cisco also emphasizes “deep buffering” as a key feature.
Rakesh Chopra , SVP & Fellow, Silicon One, commented that “deep buffers slow down AI workloads … if you dig into the details, it’s actually not true” when paired with proper congestion management.
He said the system is designed so that “everything has shrunk way down, we can drive our fan power down; we can save on power conversion … we’re chasing out every single watt in that system … because that is the fundamental limitation in the industry.” In parallel, broader reports quantify how sharply data center power demand is rising.
Goldman Sachs Research forecasts that global data center power demand will increase 165% by 2030 (relative to 2023).
Continue on AIM Media House